How Should we Visualize the Uncertainty of Political Polls?

Political polls and election forecasts are important technologies and sources of political information in modern democracies. But anecdotal evidence suggests that the inherent uncertainty of polls and forecasts is often ignored and not well understood by average citizens. This is problematic because it hampers citizens’ ability to inform their democratic choices in the best possible way. The proposed paper views this problem as a statistical communication challenge and asks how to best visualize the results of political polls to convey their uncertainty in a way that citizens easily and correctly understand. Based on recent advances in data visualization we will experimentally evaluate a series of graphical formats and techniques - variations of error bars and confidence bands, distribution plots, gradient plots, and hypothetical outcome plots – in their performance of effectively displaying uncertainty of political polls’ results. So far, very few experimental findings exist on the visualization of uncertainty in general, let alone the visualization of uncertain political information in particular. Our experimental approach is based on a full factorial of graphical formats, political information scenarios and analytic tasks administered on a) a crowd-sourced platform (such as MTurk) as well as b) in a representative online survey of real citizens. Based on these results we will derive best practices of displaying and communicating the results of political polls and election forecasts. Given the importance of the public understanding of politics for democratic governance our paper makes several important contributions. Next to catering to the exploding interest in visual methods in political science and the discipline’s “need to do a better job of data visualization” (Alvarez 2016: 15), this paper contributes to our understanding of citizens’ political sophistication by investigating how people make sense of the political world in the course of informing themselves through the media. Last, this study contributes to the important problem of publicly communicating the results of political science research – an increasingly valuable skill for political scientists to have.